In studies of traits like hypertension or obesity, information can be lost when arbitary cirteria are used to dichotomize a continous variable to determine ?affection status.? The effect of this loss of information on the power of model-independent tests of linkage and association was evaluated using computer simulation. The Genometric Analysis Simulation Program (G.A.S.P.) was used to simulate a continuous variable with a threshold set to that approximately 5% of the population would be affected. One hundred nuclear families, each with four offspring, were acertained so that at least two offspring were afected in each family. Models considered include: heritabilities from 0 to 0.8, complete and no linkage disequilibrium, and recombination fractions from 0 to 10 cM. The power of four variations of the Haseman-Elston (H-E) sib- pair test for discrete traits and the TDT test were compared to the H-E test for a continous trait. The power of the H-E test for concordantly affected pairs (only) was nearly identical to that of the H-E test for the continuous trait. The TDT test of association was more powerful than the H-E test for continuous traits when there was complete linkage disequilibrium; other the power of this test was only marginally better than the type I error rate